Thinking from Professor Yutaka Matsuo's AI Skills Lecture: The First Step to Using AI in Practice [Sharing for Small Businesses]
Watching NHK's 'Learning AI Skills from World-Class Experts: University of Tokyo Professor Yutaka Matsuo #01 ', what caught my attention wasn't the flashy AI features. It was about where to test AI in actual work and how to verify the results. When you think about that, you start to see your own way of working before you even consider how to interact with AI.
When you watch an AI lecture, you get the feeling that you could automate all your work by the next morning.
(Of course, it doesn't work out that conveniently.)
What caught my attention this time was not the flashy AI features, but the rather plain entry point of 'trying it out in your work first.' It's plain, but if you skip this, everything that follows will likely remain vague.
Hello. I am Tsuyoshi Watataru.
I run a business called Netarie, where I support small businesses in sharing content on note.
I pick up material from notes, audio, past articles, and daily insights within their work, and organize them into note articles that are searchable and convey their personality.
I also use AI in my own article production, but I've started focusing more on which processes to use it in rather than just the act of using it.
The 'try it out first' advice in Professor Yutaka Matsuo's AI lecture really stuck with me
The entry point this time is NHK's 'Learning AI Skills from World-Class Experts: University of Tokyo Professor Yutaka Matsuo #01 '.
It was a very interesting program. Professor Matsuo, a leading expert in AI research, is working hard to promote AI more widely in Japan.
The program introduced the flow of incorporating AI into actual work, such as accounting tasks for small and medium-sized enterprises, rather than keeping AI utilization as just a topic for engineers. The first step shown was 'trying to use AI in your work'.
Hearing just this, it seems obvious.
Open ChatGPT. Summarize text. Get drafts for emails. Have tables organized. If you are already doing this, you might feel, 'I've already passed that stage.'
I also thought, honestly, isn't everyone doing at least this much by now?
But 'having used AI' and 'using AI in your work' are similar, yet there is a significant distance between them.
I myself have spent a long time just playing with AI. However, even though I feel like I've done a lot of work on days when I try out new features, sometimes I find myself back to my old ways the next day.
It was convenient.
That's all.
Using AI has become the goal. There are days like that. With this, work doesn't really change much.
When you incorporate AI into your work, the ambiguous criteria for decision-making become visible.
What I find particularly interesting is not the AI itself, but the moment you try to hand over work to it.
Between humans, there are tasks that can be understood with a simple request like, 'Please summarize this nicely.' This is even more true for work you do by yourself. Even if you make decisions somewhat intuitively every time, you are able to process them in your own head.
However, when you try to hand that over to AI, you suddenly find yourself in trouble.
What should I input? How far should I go to consider it complete? What do I look at to decide that 'this is usable'?
The parts that you previously processed by intuition suddenly come to the surface in a jumble. The skill of 'verbalization' is the skill of using AI. This is a skill that AI cannot yet replace.
It is similar in article creation.
You can ask AI to 'tidy up this article.'
However, what are the words you absolutely want to keep from the original notes, where is fact-checking necessary, and what parts can be cut to make it easier to read?
If those points remain ambiguous, even if a clean piece of writing comes out, it will gradually drift away from your own writing style.
You need the skill to properly convey what you want to convey.
This is not to say that AI is bad.
Rather, it becomes a trigger to bring out the criteria you have been using for your work until now.
In small businesses, there are many judgments that only the person themselves knows, even if they aren't enough to create a business manual.
Even with a single reply to a customer, there are criteria like 'I will reply to this immediately' or 'I will check this once.' Even with articles, there is a line between 'I will publish this story' and 'I will not write this.'
When you try to introduce AI, you cannot proceed without explaining those boundaries.
There is a discovery here.
How much can you verbalize your own work? How much can you externalize it as knowledge?
How much to delegate when using AI, and where to leave human judgment
Thinking this far, what I wanted to take away from this program was not the new way of using AI itself.
When using AI, the first thing to decide is where to draw the line regarding which parts of the work should remain under human judgment.
For example, let's consider the process of creating a single article.
Picking up topics from audio or notes, organizing similar stories, and generating draft structures—these are parts where it is easy to get help from AI.
On the other hand, questions like "Is it okay to share this story now?", "Is this really what I think?", and "How much of the customer's story should I include?" are not easy to simply hand over.
I have increasingly returned to my own words at this final stage. Specifically, I leave notes via audio and have the AI read them. It's like adding flesh and blood to the polished text produced by the AI.
There are processes that can be sped up by using AI. However, just because something can be done quickly does not mean it should be left entirely to AI.
This line of "how much to delegate and where humans should judge" is not just about writing. I also thought about this same boundary in an article about AI-based health consultations.
Of course, the boundary changes depending on the job.
For tasks like routine data aggregation, where standards are clear and easy to verify, the scope that can be left to AI or automation might be wider. Conversely, for work involving customer relationships or decisions on public disclosure, there are more areas that require human oversight.
Therefore, it seems difficult to create a uniform solution when talking about incorporating AI into business operations.
Rather than thinking about large-scale AI implementation from the start, try testing one thing: "How much of this weekly task can I hand over?" At that size, it becomes easier to observe your own work.
If you think of one task you usually judge intuitively, you might find parts that are surprisingly difficult to explain.
Before handing it over to AI, try writing down just one line of your judgment criteria. There may be things that become clear from that.
Things to think about before introducing AI into your work
When you apply the points discussed so far to actual work, more detailed hesitations may arise.
You don't need to decide everything at once. It seems best to start by testing one task while putting your own judgment criteria into words.
Q. Which tasks should small companies or sole proprietors try AI with first?
It is best to start with tasks that you repeat weekly and where you can easily verify the results yourself. As mentioned in the text, if the task is small enough to consider "how much of this weekly task can I hand over?", it becomes easier to observe not only the AI but also your own workflow.
Q. How do I judge if the results from AI are "usable for work"?
Before handing it over to AI, put into your own words "what constitutes completion" and "what I look for to judge if it is usable." For an article, the words you want to keep and the parts that require fact-checking become the criteria for judgment. If you make a short note of the criteria you checked, it will also serve as a comparison for the next time you try the same task.
Q. Once a task has been successfully handled by AI, is it okay to automate it completely?
Just because something worked once doesn't mean you can immediately remove human oversight. It is safer to repeat the process first to see where potential failures might occur and at which points a human can verify the output.
I have also organized the concept of breaking down processes before fully automating with AI in another article.
Q. Should I avoid using AI for tasks that are difficult to hand over?
You don't have to hand over everything. You can let AI help with organizing or drafting, while leaving decisions—such as whether to publish or whether you truly believe in the content—to humans.
As an example of dividing roles between AI and humans, I have also organized the concept of keeping primary information in articles.
By actually trying it out once, you will start to see which parts of your work are easy to explain and which parts you are still judging based on intuition.
That note itself will likely become material for the next time you use AI.
Thank you for reading.
It's not just about using AI, but about where to use it and where humans should make the decisions.
What I was thinking about while watching this program is something I also value in my regular article production.
I have summarized how Tsuyoshi Watataru supports small business note communication here.
→ Reaching search results while conveying personality. Small business note communication support provided by "Netarie"
A list of services and the latest information are summarized on the official website.
→ Netarie Official Website
[About Production] This article is based on the structure, notes, and research of Tsuyoshi Watataru, using AI as an assistant for drafting and organizing, with final confirmation and editing done by the author himself. The angle of the article, what to keep, what to cut, and the decision to publish are all handled by Tsuyoshi Watataru.
#SmallBusiness #AIUtilization #GenerativeAI #ChatGPT #WorkSkills #BusinessImprovement #WorkDesign #AISkills #YutakaMatsuo #LearningAISkillsFromWorldClassExperts #SMEs #SoleProprietors #SoloCEO #AIandWork #BusinessDesign #WorkReview #Writing #ArticleProduction #note #noteCommunication #Netarie #TsuyoshiWatataru #Editing #ContentProduction #AIera